Drop some models/experiments

Check for better agreement after removing the major sources of variance

42 features that were were identified by variance explained were deleted from a fresh data frame and TSNE was applied to the reduced feature set. Further below those particular features are examined more closely

Only 6 features are left, suggesting many dimensions where models and data disagree

cohen score is : D = -0.15 Spikecount_stimint_1.5x

cohen score is : D = -0.64 initburst_sahp_vb_1.5x

cohen score is : D = 0.05 Spikecount_stimint_3.0x

cohen score is : D = 0.59 input_resistance

Features with high agreement between models and data

<Figure size 432x288 with 0 Axes>
-27354.21648389174

cohen score is : AP1RateOfChangePeakToTroughTest_3.0xD = -1.05

272.42181591323225

cohen score is : sag_ratio1_3.0xD = 1.17

19334.458745961303

cohen score is : ISICVTest_3.0xD = 0.34

70132.97536001515

cohen score is : ISIBurstMeanChangeTest_3.0xD = -1.03

178302578.85095984

cohen score is : peak_index_3.0xD = -0.60

178241584.288872

cohen score is : threshold_index_3.0xD = -0.60

178530810.96642792

cohen score is : fast_trough_index_3.0xD = -0.60

175527952.76589105

cohen score is : peak_index_1.5xD = -0.52

175512116.83542085

cohen score is : upstroke_index_1.5xD = -0.52

22931026.072419703

cohen score is : AP_fall_indices_1.5xD = 1.61

Features with where models and data disagree, good candidates for optimisation

<Figure size 432x288 with 0 Axes>

Least Agreement

Variance explained

0
AP1RateOfChangePeakToTroughTest_3.0x 0.208
sag_ratio1_3.0x 0.196
ISICVTest_3.0x 0.096
ISIBurstMeanChangeTest_3.0x 0.056
peak_index_3.0x 0.039
threshold_index_3.0x 0.033
fast_trough_index_3.0x 0.033
peak_index_1.5x 0.029
upstroke_index_1.5x 0.027
AP_fall_indices_1.5x 0.023

Feature extraction suite membership:

AP1RateOfChangePeakToTroughTest_3.0x belongs to Druckman

sag_ratio1_3.0x belongs to: efel

ISICVTest_3.0x belongs to Druckman

ISIBurstMeanChangeTest_3.0x belongs to Druckman

peak_index_3.0x belongs to: allen

threshold_index_3.0x belongs to: allen

fast_trough_index_3.0x belongs to: allen

peak_index_1.5x belongs to: allen

upstroke_index_1.5x belongs to: allen

AP_fall_indices_1.5x belongs to: efel

Best Agreement

Least Variance explained

0
AP_rise_time_3.0x 1.656e-03
peak_time_3.0x 1.588e-03
fast_trough_t_1.5x 1.297e-03
APlast_width_1.5x 1.163e-03
adaptation_index2_3.0x 1.002e-03
input_resistance 8.232e-04
peak_indices_3.0x 6.379e-04
Spikecount_stimint_1.5x 4.876e-04
AP_end_indices_3.0x 4.266e-04
Spikecount_stimint_3.0x 2.490e-04

Feature extraction suite membership:

AP_rise_time_3.0x belongs to: efel

peak_time_3.0x belongs to: efel

fast_trough_t_1.5x belongs to: allen

APlast_width_1.5x belongs to: efel

adaptation_index2_3.0x belongs to: efel

input_resistance belongs to Druckman

peak_indices_3.0x belongs to: efel

Spikecount_stimint_1.5x belongs to: efel

AP_end_indices_3.0x belongs to: efel

Spikecount_stimint_3.0x belongs to: efel

A good data to optimize against, as it seems closer to models. "Layer 4 aspiny 313862167", "Layer 4 spiny 479728896",